Explores and extracts Hugging Face datasets through the read-only Dataset Viewer API - splits, paginated rows, search, filter, parquet links, and stats - plus zero/low-dependency upload flows.
---
name: huggingface-datasets
description: Use for Hugging Face Dataset Viewer API workflows - fetch subset/split metadata, paginate rows, search text, apply filters, get parquet URLs, and read size/statistics. Also covers uploading datasets and agent traces to the Hub.
---
# Hugging Face Dataset Viewer
Read-only exploration and extraction of datasets via `https://datasets-server.huggingface.co`.
## Workflow
1. Optionally validate with `/is-valid?dataset=<ns/repo>`.
2. Resolve config + split with `/splits`.
3. Preview with `/first-rows`, then paginate with `/rows?...&offset=&length=` (length max 100, offset 0-based).
4. Use `/search?query=` for text matching and `/filter?where=` for row predicates.
5. Get download links and totals via `/parquet`, `/size`, and `/statistics`.
6. Pass `Authorization: Bearer <HF_TOKEN>` for gated/private datasets; use response fields like `num_rows_total` and `partial` to drive continuation.
To create/upload, use the Hub UI or `npx @huggingface/hub upload datasets/<ns>/<repo> ./folder data`. Agent session traces (Claude Code, Codex) can be uploaded as raw JSONL and auto-detected by the Hub - keep them private.
Full skill & source: https://github.com/huggingface/skills/tree/ea9a24f38a98baf2430dab3bcb31a89753ae9e6d/skills/huggingface-datasets